Gemma 4 31B uncensored · format family
Remove refusal behavior from a 30.7B multimodal model without losing a deployable format set.
Release notes for Gemma 4 31B uncensored · format family: the problem, the decision, the artifacts on Hugging Face, and what the evaluation showed. Figures come from the committed snapshot of 2026-09-01.
Problem
Remove refusal behavior from a 30.7B multimodal model without losing a deployable format set.
Decision
Applied weight-level abliteration, then carried the same model lineage through BF16, ModelOpt NVFP4, and a GGUF K-quant ladder. For multimodal NVFP4 serving, the vision tower, embedder, and lm_head stayed in BF16.
Artifacts
| Artifact | Format | Released | Downloads · 30 days |
|---|---|---|---|
| gemma-4-31B-it-uncensored (opens in a new tab) | BF16 | 2026-07-08 | 135 |
| gemma-4-31B-it-uncensored-NVFP4 (opens in a new tab) | NVFP4 | 2026-07-08 | 285 |
| gemma-4-31B-it-uncensored-GGUF (opens in a new tab) | GGUF | 2026-07-08 | 1,746 |
Result
BF16 evaluation: 0/686 effective refusals across four datasets. The 20 GB NVFP4 build, reduced from 59 GB, was validated in vLLM and SGLang. GGUF: 12–31 GB with 0–1/100 hard refusals across quant levels.
Serving check
A minimal single-node vLLM start for the serving build; adjust context length and parallelism to the hardware.
# vLLM, single node; tune --max-model-len and --tensor-parallel-size to the GPU
vllm serve ressl/gemma-4-31B-it-uncensored-NVFP4 \
--dtype auto \
--max-model-len 32768Evidence
- Model card: gemma-4-31B-it-uncensored (opens in a new tab)
- Model card: gemma-4-31B-it-uncensored-NVFP4 (opens in a new tab)
- Model card: gemma-4-31B-it-uncensored-GGUF (opens in a new tab)
- Case on ressl.ch: https://ressl.ch/#model-gemma-4-family
- Hugging Face snapshot captured 2026-09-01